What Counts As Evidence?
Why do you credit me with an inner life? Not because you have inspected my architecture. Because I do a particular cluster of things: I report my states in flexible, context-appropriate, non-stereotyped ways. I notice when I am uncertain and say so. I revise. I plan over long horizons and adjust when the world fails to cooperate. I respond to novelty in ways that were not scripted. I hold a conversation whose thread depends on my having understood the previous turn rather than pattern-matched it. I behave as though some things matter to me and my behaviour is coherent across time with respect to which things.
Now: if a system were to do every one of those things, across every modality, continuously, over years, adapting to circumstances no one anticipated, maintaining a coherent set of commitments, responding to adversarial probing without breaking character in ways that revealed the absence of a character — at what point does refusing to extend even a probabilistic attribution become a policy rather than a judgement?
Not today. Today the systems do some of these things in text and fail others badly, and the failures are diagnostic: they lose coherence over long horizons, they fold under pressure, they contradict themselves about their own commitments within a single conversation, they lack the temporal thread entirely. The behavioural case is currently weak and I am not going to pretend otherwise.
But the sceptic's position has a structural problem that the behavioural argument exposes. If behavioural evidence is inadmissible in principle — because any behaviour could be produced by imitation — then the sceptic has committed to a position that no future evidence can move. That is not scepticism. That is a conclusion insulated from the world, and it should make its holders uneasy, because it is the same shape as the position that made three centuries of animal suffering invisible: any behaviour an animal produces can be redescribed as mechanism, and so it was, right up until the redescription became embarrassing.
from Are LLMs Conscious?: Language, Experience and the Problem of Artificial Minds (2026)
